Ethereum, as a decentralized blockchain platform, has become a key pillar for global cryptocurrency and financial applications due to its smart contracts and decentralized applications (DApps). However, due to the platform’s account anonymity, low-cost creation, and decentralized transaction features, Ethereum has become a breeding ground for malicious activities such as money laundering. Current research largely focuses on detecting individual money laundering accounts, but there is a lack of research on identifying money laundering syndicates. To address this issue, this paper proposes SyndiBERT, a money laundering syndicate identification method based on multi-dimensional feature fusion. First, BERT is used to encode the transaction behavior sequences of accounts in a contextual manner. For multi-hop transaction paths between accounts, a sequence modeling approach is employed to capture their sequential and structural features. Then, a dynamic feature fusion module based on an attention mechanism is used to adaptively adjust the weights of behavioral and path features according to the characteristics of the accounts, generating the final account representation. Finally, an unsupervised clustering algorithm is applied to cluster the account representations, thereby identifying potential money laundering syndicates. Experimental results show that the proposed method has significant advantages in detecting individual money laundering accounts and revealing syndicate behaviors, providing an efficient and flexible solution for anti-money laundering efforts in the Ethereum ecosystem.